Year: 2025
Type: article
Abstract: The rapid expansion of Internet of Things (IoT) ecosystems has significantly increased the attack surface of modern networks, demanding efficient and lightweight security solutions. This paper presents a machine learning–based intrusion detection system (IDS) designed for IoT environments, focusing on the balance between detection accuracy and computational efficiency. The proposed approach employs both supervised learning algorithms (Decision Tree and Random Forest) and unsupervised techniques (K-means and DBSCAN), using the NSL-KDD dataset as the experimental benchmark. Data preprocessing includes normalization and categorical feature encoding to ensure model stability and performance. Supervised models are evaluated using accuracy and weighted F1-score, while unsupervised methods are assessed through clustering metrics such as Adjusted Rand Index, Normalized Mutual Information, and silhouette coefficient. Experimental results demonstrate that supervised models achieve near-perfect classification performance, with Random Forest slightly outperforming Decision Tree, particularly for minority attack classes. The selected model is deployed and validated on a Raspberry Pi Desktop platform, achieving high accuracy, low inference latency, and moderate resource consumption. These results confirm the feasibility of deploying intelligent intrusion detection mechanisms on edge devices, contributing to more secure and resilient IoT infrastructures.
Institution: Universidade Federal de Lavras
Language: English
Subfield: Computer Networks and Communications
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7218082801